THE BOOK--Playing The Percentages In Baseball

I like to use the replacement values provided by N Silver a couple years ago. Then, you can just add in the WPA/LI and Dewan’s fielding numbers.

If you buy Dewan’s +36 fielding runs for Pujols, then he is going to win any reasonable analysis by a good margin, I expect.

Probably a typo, but Dewan doesn’t publish +/- fielding runs, just fielding plays.

Silver’s numbers are fairly good, but his SS numbers seem out of line.  I prefer mine:
+1.0 C
+0.5 SS/CF
+0.0 2b/3b
-0.5 LF/RF
-1.0 1B
-1.5 DH

Certainly mine are easy to remember.  And 0.2 wins here or there is not going to make/break the case.

Otherwise, I agree with David generally speaking, though he didn’t talk about playing time. 

Basically calculate WAR, using WPA/LI as your base hitting numbers, UZR/Dewan/Fans for your fielding, and the above positional adjustments.

You can arguably include the player’s clutch score.

***

As it pertains to Jimmy Rollins, Dewan’s leaders in SS from 2005-2007 are:
+92 Everett
+45 Bartlett
+43 Barmes
+42 ROLLINS
+41 Jack Wilson
+36 Furcal

If someone wanted to say that Jimmy Rollins was one of the best fielding SS in the NL in 2007 (Everett missed most of the season), even if his sample fielding numbers of 2007 were not that good, it’d be hard to argue against.

One standard deviation, just by luck, must be somewhere around 8 or 10 runs.  So, a guy can be a true +10 runs, and 95% of the time, he’ll put up -10 to +30 just by luck in any given season!  (It’s like crowning an OBP champ after two months of play, as if that specifies how good a hitter he is at getting on base.) Clearly, one year of fielding stats is pretty much cr-p.  As we saw with the forecasting analysis with Marcel, when I have to regress at least 50% or more for the hitting stats, I may as well have guessed league average for all the good it did me.

For the 2005-2007 time period, Dewan loves Jimmy Rollins (+10 runs per year).  The Fans love Jimmy Rollins’ glove (+14 runs). 

From 2003-2007, UZR sees Rollins as being +4 per 162 G (+6 from 2003-2006, -6 in 2007).  Clearly, if someone wanted to call Rollins a true +4, where he was a bit lucky in 03-06 and very unlucky in 07, that’s ok, as per UZR.

So, if you look at Dewan’s numbers, and figure in 2007 he was a bit worst than his 05-07 average (say +8 runs), and you look at the Fans (+14, and maybe they are a bit biased, so make it +10) and look at UZR (average +4, so give him +2 for aging), you can reasonably call Rollins’ fielding a +7 in 2007.

You don’t want to know what UZR/Dewan thinks of Hanley’s fielding… ok, around -20 runs.  Fans aren’t as harsh (-10 runs).

So, the gap in their fielding can reasonably be argued at 25 runs, which is about their gap in their hitting. 

FWIW, Hanley had a horrible clutch year.

---"his SS number seems out of line”

That’s the point of using his numbers. I wasn’t aware that Silver’s SS number has been refuted. Certainly, he comes by his numbers in a different way than you do yours. I haven’t thought about which one is more appropriate for an MVP discussion.

As far as using 2005-2007 fielding numbers for Rollins, the problem there is that if Rollins numbers are worse for 2007, it doesn’t have to be due to the sample size, it could be that he really fielded worse. I can see combining Dewan, MGL, Pinto, etc., for 2007. But using other seasons in an MVP discussion?

They could all be biased in the same way.  So, if his fielding numbers have a larger amount of uncertainty than his hitting numbers, then you can’t weight them at 100% each.  Suppose all you had were the Palmer fielding runs and the Palmer hitting runs.  There’s simply no way you would add the two to get an overall run value, because the uncertainty in one measure is far higher than the other.  We know a guy hit 35 HR.  We don’t know that a guy saved 23 singles.  We’re making a guess, with a certain amount of uncertainty.

One way to account for that is to: look at prior years of data of performance and look at what the Fans saw.  For example, Betancourt showed a noticeable dropoff in performance according to the Fans.  Therefore, by accounting for his performance prior to 2007, and combining it with how the Fans see him, this reduces your uncertainty of your 2007-only measure, if you adjust it accordingly.

Hi, thanks for the feedback.

Re: missing fielding. I should have been clearer about that in my post on the BTF thread. I certainly didn’t intend for the numbers to be taken as total contribution, just hitting (and baserunning) relative to positional average.

Tango, let me see if I understand what you are saying here:

From 2003-2007, UZR sees Rollins as being +4 per 162 G (+6 from 2003-2006, -6 in 2007).  Clearly, if someone wanted to call Rollins a true +4, where he was a bit lucky in 03-06 and very unlucky in 07, that’s ok, as per UZR.

UZR has Rollins as follows per 150 games,

2003 +9
2004 +6
2005 +7
2006 +3
2007 -6 (per 162)

So, 2007 is the outlier. Two ways that could have happened: he had an off year defensively and the UZR reflects that, or the balls hit to him were misclassified by the formula as being easier to field than they actually were, and so his rating gets unfairly dinged.

At this point MGL would say the most likely possibility is that both things happened to some degree. In other words, he fielded slightly worse, and his batted balls were slightly more difficult in a way that UZR didn’t notice.

Let’s say that if you fit the UZR data to a normal aging curve, you get:

2003 +5
2004 +5
2005 +4
2006 +3
2007 +2

Then the guess for 2007 should be between +2 and -6 (before we look at Dewan and the Fans). In this case we can get away with just splitting the difference, but there are fielders with much wider y-t-y swings. What I want to know is, is there a general way to deal with this question?

I suppose the necessary information is, first, the year-to-year variability in fielding performance, and second, the relative accuracy of UZR. By that I mean a formula saying, if player X had a UZR of +5 runs over a full season, here is the percent chance that his true performance was in the range +0 to +10.

Do we have that kind of formula? Any ideas on how to calculate it, maybe by comparing against other rating systems? (But what if they all miss in the same way?)

As for refuting Silver, didn’t we talk about this in another thread?  IIRC, his comparison group was biased because he selected players that were at least 27 (and paid close to league minimum).  By that age, alot of guys who would qualify at SS were moved over to 2B or 3B.

And when I repeated that methodologies for pitchers, I ended up with replacement pitchers with a 4.40 ERA.

Ah, you all beat me to the punch on the fielding reliability issue. I guess eyeballing it (as proposed in #5) will work most of the time, but it would be good to get some quantitative information on reliability.

On a different note, I’m unsure about WPA/LI as a good stat for MVP consideration. Say one guy hits a HR with the bases empty and 2 outs, and another guy hits a HR with men on 2nd and 3rd and nobody out. In both situations it is a tie game in the 1st inning. If I’m not mistaken, the first guy will have higher WPA/LI. Is that the way we want to give credit?

My real issue is, WPA/LI assigns equal importance to every PA. Why is it better than straight WPA when you want to determine the MVP?

On a different note, I’m unsure about WPA/LI as a good stat for MVP consideration. Say one guy hits a HR with the bases empty and 2 outs, and another guy hits a HR with men on 2nd and 3rd and nobody out. In both situations it is a tie game in the 1st inning. If I’m not mistaken, the first guy will have higher WPA/LI. Is that the way we want to give credit?

Interesting question.  I believe that both events are about equal from a BRAA and Linear Weights perspective.  In WPA terms, the first event has a WPA of .102 with an LI of 0.4, while the second event has a WPA of .132 with an LI of 1.9.  So there is a big difference between WPA and WPA/LI.

My own take for this specific situation is that straight WPA and/or BRAA or LW are best—I think the hits are about equal in impact.

I have a feeling we could go all day finding examples like that one, and debating “what’s best.” Might be fun.

Another example: guy hits a home run in a tie game with no one on and two out in the bottom of the ninth.  WPA of .466 and LI of 1.2.  WPA/LI of .388.

The home run with two out and bottom of the first, no one on, tie game, has a WPA of .104 and LI of .4.  WPA/LI of .255.

This is the sort of difference that WPA/LI was crafted to address—for those who don’t think a home run in the ninth should be THAT much more valuable than a home run in the first.

Going to fret’s example of hitting a HR with 2 outs and bases empty: the HR here is HUGE!  The reason is that you have such a little chance of scoring in this inning, that you NEED to change your hitting approach such that you are focusing on power.

Look here:
http://tangotiger.net/RE9902event.html

A single or walk gives you just +0.13 runs, which a HR gives you 1.00.  This is like Pedro on the mound in his heyday (or Mo)… getting singles off him aren’t going to hurt him.  A HR will kill him.

Now, the other one is getting a HR with 2 guys in scoring position and no outs.  A HR is almost overkill here.  You want a Wade Boggs here, not a Jim Rice.

So, WPA/LI describes, in real baseball terms, exactly how you need to approach the PA.

Just wondering, what is the evidence that WPA/LI is a skill?

---"They could all be biased in the same direction”

It is an interesting problem how to use multi-year data, plus regression to the mean, to arrive at an estimate of actual value in year x, instead of true talent.

Okay, so I struggle to articulate what WPA/LI means.  Here’s an attempt:

- BRAA: Run impact of an event based on the actual base/out situation in which each event occurred.

- WPA/LI: Game impact of an event based on the actual base/out/score situation in which each event occurred (IOW, without the “time” element).

- WPA: Game impact of an event based on the actual base/out/score/inning situation in which each event occurred (time element added in).

You get to WPA/LI through WPA.  Dividing by LI basically takes the clock out of the equation.  So, if you trust the inherent logic behind WPA (regardless of whether you “buy into it” or not), you should trust the logic behind WPA/LI.

If that’s right, then, the interesting discussion for each situation (as Tango provided in his post) is WHY WPA/LI varies from BRAA (the variance from WPA is clear).

Tango, have you ever looked at which situations and events yield the biggest differences between BRAA and WPA/LI, holding the score differential constant (IOW, always looking at tie games or one-run games)?

Studes, I think you are close, but not all the way there.

Let’s focus just on straight Linear Weights, where the run value of a HR is 1.40 (and the win value of a HR is around 0.13 wins).

How did we get the run value?  That’s simple: the run expectancy when a HR is hit is around 0.533 runs before the HR, and 1.942 runs after the HR.  That difference is 1.409 runs.  (All that is from The Book, table 4, p.21.)

We do the EXACT same thing for the win value of a HR (Table 11, p.43): the win expectancy when a HR was hit was .514 before the HR was hit, and .638 after the HR was hit.  That difference is .123 wins.

The WPA is +.123 and the LWTS is +1.409.  These two numbers are calculated EXACTLY the same way.

You could in fact then figure out the LWTS HR run value by game state: how many runs are added by inning,score,base,out.  Now, clearly, the inning and score don’t impact how many runs are added: only the base and out.  You don’t get more runs if you are in the 1st or 7th inning.  That’s why BRAA is enough: Linear Weights by the 24 base out states.

AND, we didn’t have to use runs, we could have used wins.  But, since the inning and score don’t impact the equation, there’s no point to going to the trouble of knowing the change in win expectancy.  You only have the 24 base out states.  You can then do a quick conversion to wins.  You should still however, do BRAA/LI, to remove the impact of the leverage, otherwise a guy with a high LI (base/out state of lots of runners on base) will get a boost he doesn’t deserve.

For WPA/LI, you are still doing the WPA part.  But, like I said, the WPA part you are doing ALREADY for Linear Weights (runs).  It’s the exact same thing.  The wrinkle is that the context is that the player had a disproportionate number of game states in high or low leverage situations.  So, you need to depress that out.

WPA/LI gives you that.

As I mentioned elsewhere, if the bases are loaded in the bottom of the 9th of a tie game, the wOBA equation would be:
wOBA = (bb+1b+2b+3b+hr)/PA

That is, a walk is as good as a HR.  WPA/LI is that equation.  It’s exactly that.

WPA/LI is wOBA by the game state (inning,score,base,out).

Sorry, but I just don’t get it.  Your last sentence, “WPA/LI is wOBA by the game state (inning,score,base,out)” is utterly confusing to me.

Also, WPA and BRAA are not the exact same thing.  They’re the exact same methodology, but the underlying values are different.

If what you’re saying is true, then why do I get the exact same figure for WPA/LI when I take a specific situation and outcome and only vary the inning?  That is, WPA/LI for a two-run double in the first is the same as it is in the ninth (given the same preceding situation and score differential) and every other inning as well.

Ignore that second paragraph.  I see that when you said “exact same thing” you were just following up on your earlier point.  Context!

The double is a very neutral type of event.  It will always be worth somewhere between a HR and a single.  The HR and the walk are the ones whose win values will have the most change in impact based on the game situation.

If you look at WPA/LI for HR,3b,2b,1b,bb in the first and ninth innings of your situation, report back the numbers, and you’ll see that kind of convergence.

***

And remember how to do a wOBA equation:
1. add the run or win value of the out to each of the positive events
2. scale the coefficients of the positive events upwards, such that the coefficient times the frequency equals exactly the sum of the frequency of the positive events (i.e., OBP)

There is a relationship between wOBA and WPA/LI.

... for each game state.

I looked at bases-loaded walks with two outs in the bottom of each inning with the scored tied.  Each WPA/LI came out to approximately .06.

I’m using my WPA spreadsheet, which has the “short” version of your LI formula.

Maybe I’m wrong, but if LI is based on four factors and you change only one of those factors across different cases, won’t that make WPA/LI equal in each case?

Studes, I’ll have to take a look (at the WE/LI file I sent you, rather than your WPA sheet).  It would have to be on an upward slope toward around +.055 wins or so in the 9th inning.  I certainly wouldn’t have expected it to be fairly static around there.  Doesn’t make any sense frankly.

After all, the HR value in the 9th inning would also be EXACTLY the same as the walk in the 9th inning.  It cannot be the same in the 1st inning.  So, if the HR value is higher, then the walk value must be lower, since the average of all positive events has to be +.055.

Studes, this is what I get, when I have bases loaded, 2 outs, in the bottom of each inning:

Inn WPA LI WPA/LI
1 0.090 2.70 0.033
2 0.096 2.87 0.033
3 0.103 3.06 0.034
4 0.113 3.31 0.034
5 0.127 3.61 0.035
6 0.147 4.00 0.037
7 0.179 4.53 0.039
8 0.234 5.28 0.044
9 0.336 6.39 0.053

This makes perfect sense.  The random win value of a walk is +.03 wins.  The maximum win value of a walk is the minimum win value of a HR, and that’s roughly +.054 wins.

(Where’d I get that .054 value?  Imagine baseball where you have either a hit or out.  Two times out of three, you have an out, and the win value of the out is -.027 wins.  In order to balance it out, the one time in three that you get a hit, it must be worth +.054 wins.  And the only time you can have a situation like that is when the BB=1B=HR, and that happens in the bottom of the 9th, bases loaded, tie game.)

Anyway, that progression explains exactly how a hitter should gear himself towards getting more walks, as its impact, relative to the other positive events goes up.  Remember, what we are doing is centering each positive event, around the average of all the positive events.  And the average of all the positive events is exactly equal at every single game state in baseball.

It’s for this reason that WPA/LI is the single best stat you want.

Thanks, Tango.  Obviously, I have to go back and check my spreadsheet.  WPA/LI still isn’t intuitive to me, and I have no idea how to explain its nuances beyond a simple definition ("normalized WPA").  But I’ll keep thinking about it.

When you say this…

And the average of all the positive events is exactly equal at every single game state in baseball.

You’re talking about WPA/LI, not WPA, right?

Right.

Specifically, I mean that if you take the WPA/LI value of each event (say .03 for the walk, .05 for the single, .13 for the HR), and multiply those by the frequency of the walk, single, HR, you will get a value that is exactly equal to the win value of the out times the frequency of the out.

WPA/LI keeps the plus side of the ledger fairly equal to the minus side of the ledger.

The average negative event is roughly equal to -.027 wins for any game state (i.e.,take WPA for the out and divide by LI of that game state, and you’ll get around -.027 wins).

The average positive event is roughly equal to +.054 wins for any game state (WPA/LI).

It’s just a matter of how much the walk and HR et al are worth around that +.054 win value. Sometimes the HR will be worth +.11 wins, other times +.19 wins, etc, etc.  It’s all based on how valuable a HR is, relative to a walk or single.

And here’s the win value of the out, in the same bases loaded, tie game, 2 out situation:

Inn WPA LI WPA/LI
1 (0.071) 2.70 (0.0262)
2 (0.075) 2.87 (0.0261)
3 (0.080) 3.06 (0.0262)
4 (0.086) 3.31 (0.0261)
5 (0.094) 3.61 (0.0261)
6 (0.104) 4.00 (0.0261)
7 (0.118) 4.53 (0.0260)
8 (0.137) 5.28 (0.0259)
9 (0.164) 6.39 (0.0257)

As you can see, exactly as I said: the WPA/LI value of the out is constant.  (Would have been exactly the same if my WPA and LI numbers themselves weren’t rounded.)

See, while the WPA value of the out gets more impactful as you go on later in the game, it’s impact is EXACTLY of the same magnitude as the LI.  This is the entire point of the relationship of WPA and LI.  Leverage Index is the potential swing in win expectancy.  So, if the out has more swing on the negative side, then the hr, hits and walks has more swing on the positive side to the exact same extent, and LI captures that.

And so, WPA/LI removes that impactful value, and simply looks at it in terms of the relative impact of the HR, hit, and walk to themselves and to the out.

And here’s the bases EMPTY two out walk, tie score, bottom half of each inning:

Inn WPA LI WPA/LI
1 0.034 0.87 0.039
2 0.036 0.92 0.039
3 0.039 0.99 0.039
4 0.042 1.07 0.039
5 0.045 1.17 0.038
6 0.050 1.32 0.038
7 0.055 1.52 0.036
8 0.063 1.82 0.035
9 0.073 2.31 0.032

As you can see, while the WPA value of the walk always increases (from .034 wins to .073 wins), the LI increases at a faster rate, such that the WPA/LI win value of the out DECREASES.  That is, while you can get away with a walk in the 1st inning of a 2 outs bases empty situation, it is not as profitable in the 9th inning.  At that point, it’s more important to get into scoring position (or a HR of course).

So, the win value of the walk has been rebalanced based exactly on the game conditions.  A smart pitcher will pitch to that knowledge (that a walk isn’t as costly in the 9th inning as the 1st inning) and the batter will swing to that knowledge (he should go less for a walk and more for a hit).

And, as a baseball fan, you have that kind of feeling anyway.  WPA/LI simply quantifies it for you, how much the balances has shifted.

Bases empty, 0 outs, down by 2, hit a HR:

Inn WPA LI WPA/LI
1 0.092 0.91 0.101
2 0.098 0.97 0.101
3 0.105 1.05 0.100
4 0.114 1.14 0.100
5 0.124 1.26 0.098
6 0.134 1.41 0.095
7 0.144 1.59 0.090
8 0.144 1.80 0.080
9 0.116 1.98 0.058

As you can see, by WPA, it has more impact in the 7th or 8th inning.

But, “situational hitting” wise, if you do it in the 9th inning, it’s not that much different from a typical positive event.  Like I said, your typical positive event has a win value of around +.054.  It goes to show how tough it is to come back with even just a 1-run deficit.

I think I’m going to call WPA/LI as Situational Wins.

This is helpful, Tango.  Thanks.

Tango #28 - This chart is why I don’t particularly care for WPA/LI.  Anybody who hits a home run with the bases empty and no outs has done everything that he can do as a batter to help his team win the game no matter what inning he does it in.  He shouldn’t be penalized by receiving only 58% of its value because he does it in the ninth rather than in the first four innings.  There is no evidence to show that either the pitcher or the batter are approaching these at bats any differently because of the inning in which they occur.

Ah, but it’s up to you to show that however.

After all, it is without a doubt that with runner on 3B and less than 2 outs that a pitcher would much prefer a K than another type of out, and will therefore pitch to that.  And, a batter will want to avoid a K as much as possible and will therefore bat against that.  Basically, they are both trying to do the opposite of what the other guy wants.  The league mean will basically not show any difference, but the standard deviation, I’d contend, must be larger of K/PA with man on 3B and less than 2 outs, than otherwise.

And this is the case for all game situations.  Everything is about situational hitting to some extent or other.

Even consider the most basic of situational hitting: the 3-0 count and the 0-2 count.  The heat zones that have been recently produced show no doubt at all that pitchers will throw to a different part of the strike zone, and different pitches based on the 3-0 and 0-2 count.  And, the batter will also swing a different way.  This is logical and expected.  Now we know the extent of it.

I don’t think you can then suppose that situational hitting can only exist at the count level, but not at the game state level.  That I haven’t shown it doesn’t mean that the status quo is that it doesn’t exist.  Logically, players respond the approach to the context.  (The extent of which, we don’t know.  And it’s only on that basis, that you could argue and agree with me, that perhaps the WPA/LI should be different, but nowhere near as what my charts above are showing.)

In my opinion, anyway.

By “these at bats” in my previous post I was refering specifically to nobody on nobody out down by two in the late innings as opposed to the early innings.  There is no logic that the batter would approach those at bats any differently or the pitcher for that matter. 

There is certainly situational hitting at the count level and at the base out level.  If one wanted to award value at the base out level you need only to give the batter the change in run value added from the RE table instead of his linear weight.  This was something I argued for 21 years ago when I presented the first RE tables from a years PBP data at the 1987 SABR conference.  But it is only useful for evaluating past performance. There is no evidence that it is a repeatable skill and therefore is not as useful as linear weights for projection purposes.

I have no doubt that there are instances where the batters change their approach to hitting due to the game score, but in my opinion they are very rare.  They would only occur when the game winning value of scoring a single run is clearly greater than the game winning value of scoring multiple runs.  And the value of WPA/LI in those few instances would be limited to evaluating past events unless you have evidence that batters have a repeatable skill of successfully altering their approach to hitting.  If you don’t then WPA/LI is useless for projection purposes.

The other problem is that WPA, even when adjusted for run environment, is too generic a stat to represent the true chances of winning a game for a specific situation.  You would need a very detailed sim that includes full pitching and hitting data for each player to give you the actual win percentages.

Excellent point about using th generic WPA. After all, with the young Pedro on the mound, you are basically needing to play small ball.  SB breakeven% go down, the sac bunt makes more sense, the win value of a walk goes down, and relatively speaking, the HR is far more powerful.

It’s certainly possible that the nuances in changes in win value of say the walk is too small for the batter too detect, and therefore, he won’t change his approach.  Unlike say the man on 3b, less than 2 out scenario, where the win and run value of the K changes drastically.

Same with runner on 1b and less than 2 outs, where the win value of the GB also changes drastically.  Not to mention that the batter and pitcher are both conscious of the effect of the GB, they are both working with a different fielding alignment.

Just as the wall at Fenway will affect the batter/pitcher because of its drastic nature, so will anything of that sort that is noticeable.

It’s not just the skill of the pitchers that affects WPA and makes WPA/LI auapect, it is also batting order position.  A walk to a number 1 hitter has a much greater chance of scoring than a walk to a number 5 hitter.  And in the late innings the quality of the bench players available to pinch hit affects the win probability as does the quality of the relief pitchers available.

In your example it is not that the “nuances in changes of win value of a walk are too small to detect.” Any reasonably smart batter knows that when he is batting with nobody on and nobody out and behind two runs that “a walk is as good as a hit” and he needs to get on base any way possible.  He may even be aware of the nuance that it is even more important in the late innings.  But a batter can’t walk on a pitch in the strike zone, and it does him no good to let a strike go by if it is one he feels he can hit.  Once he has committed to a swing, his job is to hit the ball HARD somewhere, because that gives him the best chance to get on base.  If it happens to go over the fence, so much the better.  He shouldn’t receive partial value for that outcome just because the inning is the eighth instead of the third.

He should, if in that case a triple is as good as a HR.

Take the egregious case of hitting a grand slam of a tie game in the bottom of the 9th inning (what ARod did last year).  A pitcher doesn’t care if it’s a HR or a walk.  It’s the same outcome.  So, he’s not trying to prevent the HR at the expense of getting more walks: he’s just trying to minimize OBP as much as possible, and SLG be damned.

A similar situation is with bases loaded and 0 outs: the win impact of a triple is virtually identical to that of a HR.  So, why should we count the HR as being +.13 wins and the triple at +.10 wins?  (Not looking at inning/score, jsut base/out.) That is not what happened.  In this case, the pitcher knows that any extrabase hit is going to kill him, and he’s trying to get an out, or just allow singles.  There’s a shift in the balance.

I think we both see the same thing, but we’re disagreeing as to what it really means.

My last paragraph was just referring to the 0 on 0 out behind by two situation.  But lets take your bases loaded 0 outs situation.  Using 2005-2007 RE table 1230 base out is 2.39.  30 ( after a triple) baseout is 1.47.  RVA for a triple in that situation is 2.08.  RVA for a home run is (4 +.54) - 2.39 or 2.15.  Win value at 10 runs per win is .215 for a HR, .208 for a triple or virtually identical.  RVA handles situational hitting for baseout situations just fine.  But there is no predictive value unless you can show that more than average triples are hit in that situation than HRs.

Certainly I agree that RE by the 24 base/out states perfectly captures the run impact of the triple vis-a-vis the HR.  It’s on that basis that I had the thought (almost 90% of the time, the guy from 3B will score with 0 outs, compared to the obvious 100% for the HR; and both will drive in the same number of runners).

Lots of stuff being written (here and the other threads), so maybe I’m not keeping pace as well as I should.

Are we agreeing that, at least on a base/out level, that situational hitting exists, and can be evaluated to that extent?

Tango/35: Minor point, but when Rodriguez hit the grand slam in bottom of the ninth with two outs, they were losing 7-6. So a walk and HR would have different outcomes in that example. Doesn’t change your point about the tie game situation, of course.

I believe that on a baseout level that situational PITCHING exists.  For the most part the hitter can only react to where the pitcher throws the ball.  The only value that I see in evaluating situational hitting at any level is for career numbers after a career is over or for awarding the MVP if a hitter performed significantly better than his linear weights would indicate in high value situations.  I also think there are significant problems in using any methodology to evaluate situational hitting.

To the extent that situational pitching exists, situational hitting exists as well.

For example, let’s say you have Fenway’s wall.  Let’s also say that hitter do NOT change the way they bat.  However, certain families of hitters will end up with different outcomes because of the wall.  Even if say Wade Boggs and Fred Lynn and Jim Rice did not consciously change their approach, or even if they did not change it at all, their particular skillsets would have allowed them to exploit something that was in their environment.

The same idea applies to situational pitching.  Clearly, we have the man on 3b, less than 2 outs, where all pitchers would love the K.  So, pitchers change their approach.  Even if hitters do not change their approach in kind, or at all, certain families of hitters will be more hurt by this situation.  Those batters need to be evaluated in that environment.

To the extent that a new environment has been presented itself to the players, they need to be evaluated against that new environment.  And by environment, I mean something real, like fences, pitcher changing approach.  The inning/score may not be “real”, so I can grant that we don’t (necessarily) want to go that far.